CN111919128B

CN111919128B - Method for estimating state of charge of power storage device and system for estimating state of charge of power storage device

Info

Publication number: CN111919128B
Application number: CN201980021051.3A
Authority: CN
Inventors: 伊佐敏行; 千田章裕
Original assignee: Semiconductor Energy Laboratory Co Ltd
Current assignee: Semiconductor Energy Laboratory Co Ltd
Priority date: 2018-04-06
Filing date: 2019-03-29
Publication date: 2024-09-17
Anticipated expiration: 2039-03-29
Also published as: JP7446990B2; CN111919128A; JPWO2019193471A1; WO2019193471A1

Abstract

The present invention provides a method for estimating a state of charge of a secondary battery with high accuracy even if the secondary battery is degraded. The present invention also provides a system for measuring the capacity of a secondary battery, which estimates the SOC in a short time at low cost and high accuracy. A neural network is used in order to further improve the estimation accuracy of the SOC obtained by performing calculation processing using a regression model such as kalman filtering. The data obtained by the optimization algorithm is used as the supervision data in the neural network, and the SOC can be estimated with high accuracy by the initial parameters obtained by using the neural network.

Description

Method for estimating state of charge of power storage device and system for estimating state of charge of power storage device

Technical Field

One embodiment of the present invention relates to an article, method, or method of manufacture. In addition, one embodiment of the present invention relates to a process, machine, product, or composition (composition of matter). One embodiment of the present invention relates to a semiconductor device, a display device, a light-emitting device, a power storage device, a lighting device, or an electronic apparatus. Further, one embodiment of the present invention relates to a method for estimating a state of charge of an electric storage device and a method for controlling charge of the electric storage device. And more particularly to a state of charge estimation system of an electric storage device, a charging system of an electric storage device, and a control system of an electric storage device (also referred to as BMS (Battery MANAGEMENT SYSTEM: battery management unit)).

Note that in this specification, the power storage device refers to all elements and devices having a power storage function. Examples thereof include secondary batteries (also referred to as secondary batteries) such as lithium ion secondary batteries, lithium ion capacitors, nickel hydrogen batteries, all-solid-state batteries, and electric double layer capacitors.

Another embodiment of the present invention relates to a neural network and a control device for an electric storage device using the neural network. Further, one embodiment of the present invention relates to a vehicle having a BMS using a neural network. In addition, one embodiment of the present invention relates to an electronic device using the neural network. The present invention relates to a power storage device that is applicable not only to a vehicle but also to a power storage device that stores electric power obtained from a power generation facility such as a solar power generation panel provided in a structure or the like.

Background

As methods for estimating the remaining capacity of the secondary battery, there are coulomb counting method and OCV (Open Circuit Voltage) method.

When the conventional method is used for a long period of time and charging or discharging is repeated, errors accumulate, and the estimation accuracy of the SOC (State of Charge) may be significantly reduced. Further, since the initial SOC (0) is changed due to self-discharge even when the battery is not in use, it is difficult to improve the estimation accuracy of the SOC. The coulomb counting method has a disadvantage that an error of the initial SOC (0) or an error of the accumulated current sensor cannot be modified. Patent document 1 discloses a technique for estimating the state of a secondary battery at a low temperature with high accuracy by a state estimating unit based on information including parameters related to temperature.

[ Prior Art literature ]

[ Patent literature ]

[ Patent document 1] Japanese patent application laid-open No. 2016-80693

Disclosure of Invention

Technical problem to be solved by the invention

The vehicle mounted with the secondary battery can charge the secondary battery with regenerative electric power generated at the time of braking or the like, and there is a possibility that the secondary battery cannot be used appropriately due to overcharge. In order to prevent the occurrence of overcharge and overdischarge in advance, it is necessary to estimate the remaining charge of the secondary battery, that is, the SOC of the secondary battery with high accuracy. The invention provides a method for estimating a state of charge of a secondary battery or a method for controlling an electric storage device with high estimation accuracy.

In the case of secondary batteries, small individual differences may occur even in the same manufacturing lot due to small differences in the amount of active material, electrode size, and the like during the assembly. In a vehicle or the like, a plurality of secondary batteries are used, and when the plurality of batteries are combined, there is a case where a difference in capacity between vehicles increases due to degradation.

In addition, as the deterioration of the secondary battery progresses, the estimation accuracy of the SOC may be significantly reduced. Further, the SOC is defined according to a ratio of the remaining amount with respect to the maximum capacity of the secondary battery. When the secondary battery is discharged after full charge and the maximum capacity of the secondary battery is calculated from the time integral of the current, there is a concern that the discharge takes a long time.

The present invention provides a method for estimating a state of charge of a secondary battery with high estimation accuracy even if the secondary battery is degraded. The present invention also provides a system for measuring the capacity of a secondary battery, which estimates the SOC in a short time at low cost and high accuracy.

Means for solving the technical problems

The charge state estimation method of the secondary battery disclosed in the present specification uses a neural network in order to further improve the estimation accuracy of the SOC obtained by performing calculation processing using a regression model such as kalman filtering. The state of charge (SOC) is estimated by using artificial intelligence (AI: ARTIFICIAL INTELLIGENCE) such as a neural network.

The structure of the invention disclosed in the present specification is a method for estimating a state of charge of an electric storage device, comprising the steps of: determining a circuit model of the power storage device; in a circuit model (a foster circuit model) of the power storage device, a current is taken as an input and a voltage is taken as an output; optimizing to reduce an output error of a voltage of the power storage device, and calculating an initial parameter (first value) of a circuit model of the power storage device; storing initial parameter groups corresponding to input values of different currents; an initial parameter (second value) is determined by neural network processing using the initial parameter group obtained by optimization as the supervision data, and the initial parameter (second value) is used by kalman filter processing to estimate the state of charge (SOC).

The initial parameter group corresponding to the input value of the different current does not have to use the actually measured cycle characteristics, and may be generated based on the type setting conditions of the power storage device used by the practitioner. Another aspect of the present invention is a method for estimating a state of charge of an electric storage device, including: determining a circuit model of the power storage device; in a circuit model of the power storage device, a current is taken as an input and a voltage is taken as an output; optimizing to reduce an output error of a voltage of the power storage device, and calculating an initial parameter (first value) of a circuit model of the power storage device; generating an initial parameter group different from the initial parameter; the initial parameter group is used as the supervision data, and the initial parameter (second value) is determined by the neural network processing, and the initial parameter is used to estimate the state of charge (SOC) by the kalman filter processing.

Kalman filtering is one of infinite impulse response filtering. In addition, multiple regression analysis is one of the multivariate analyses in which the independent variables of the regression analysis are plural. As the multiple regression analysis, there is a least square method or the like. While regression analysis requires a large number of time series of observations, kalman filtering has the advantage that the most suitable correction coefficients can be obtained gradually by accumulating a certain amount of data. Furthermore, the kalman filter may also be applied to non-stationary time sequences.

As a method of estimating the internal resistance and the state of charge (SOC) of the secondary battery, nonlinear kalman filtering (specifically, lossless kalman filtering (also referred to as UKF)) may be used. Furthermore, extended kalman filtering (also known as EKF) may also be used.

Initial parameters obtained by an optimization algorithm are collected every n (n is an integer, for example, fifty) cycles, and these data groups are used for the neural network processing of the monitor data, whereby estimation of the SOC with high accuracy can be performed.

The learning system comprises a supervision generating device and a learning device. The supervision data generating device generates supervision data to be used when the learning device learns. The supervision data includes evaluation of the tag corresponding to the data of the same processing object data as the identification object. The supervision data generating device includes an input data acquiring unit, an evaluation acquiring unit, and a supervision data generating unit. The input data acquisition unit may acquire data from data stored in the storage device, or may acquire input data for learning through the internet, the input data being data for learning, and including a current value or a voltage value of the secondary battery. The supervision data may not be actually measured data. For example, by diversifying initial parameters according to condition settings, generating near-actual measurement data and using these predetermined characteristic databases for supervision data to perform neural network processing, the state of charge (SOC) can be estimated. The SOC estimation of the same type of battery can be efficiently performed by generating near-actual measurement data from the charge/discharge characteristics of one battery and performing neural network processing using these predetermined characteristic databases for the supervision data.

In the case of SOC estimation using only the optimization algorithm, the calculation amount of the optimization algorithm is large, and there is a problem that convergence to a meaningless value or divergence from an optimal value cannot be determined. The characteristics of the battery are nonlinear, and five initial parameters are obtained by a numerical optimization method of a nonlinear function. The five initial parameters are the total capacity FCC (Full CHARGE CAPACITY), the direct current resistance R _s(R₀), the resistance R _d resulting from the diffusion process, the diffusion capacity C _d, and the initial SOC (0). Note that FCC (also referred to as full charge capacity, total capacity) is a rated capacity at normal temperature of 25 ℃.

The optimization process for obtaining five initial parameters may be performed by using means mounted on Python (registered trademark) or Matlab (Matrix Laboratory: matrix laboratory) (registered trademark).

When deterioration of the secondary battery progresses, there is a concern that an error in SOC will occur when the FCC of the initial parameter is greatly changed, and therefore the initial parameter used for estimating the calculation of SOC can be updated. The updated initial parameters are calculated by an optimization algorithm using data of charge and discharge characteristics measured in advance. By performing calculation processing using a regression model of the updated initial parameters, for example, kalman filtering, it is possible to estimate SOC with high accuracy even after degradation. In this specification, the calculation processing by using the kalman filter will also be referred to as kalman filter processing.

The timing of updating the initial parameters may be arbitrary, but in order to estimate the SOC with high accuracy, it is preferable to update the initial parameters periodically and continuously as the update frequency increases.

The step of estimating the SOC will be described in more detail below.

In the first stage, a voltage value or a current value of the secondary battery is measured by a measuring means (voltage detecting circuit or current detecting circuit). These data are acquired by a voltage measurer or a current measurer (also referred to as a current sensor) and stored in a storage device. The initial SOC (0) is calculated from the voltage value obtained by using the voltage measuring instrument, specifically, from the charge/discharge characteristic data. The initial SOC (0) is an initial value of SOC. The initial Rs is an initial value of the dc resistance Rs (also referred to as R ₀), and is a resistance generated by the electrophoresis process of the ions. Five initial parameters, specifically initial SOC (0), FCC, R ₀、R_d, and C _d, can be obtained from the pre-measured charge-discharge characteristics by using an optimization algorithm, specifically by using the Nelder-Mead algorithm. Note that the Nelder-Mead algorithm is an algorithm that does not require a derivative function.

As another method for calculating the initial SOC (0), the initial SOC (0) may be determined by measuring the open circuit voltage of the battery before the start of use by using a voltage detection circuit and using a map or a correspondence table of the open circuit terminal voltage OCV and the SOC obtained in advance. OCV is the voltage of a battery in an electrochemically balanced state, and corresponds to SOC.

And constructing a fully-connected neural network in the second stage. The charge voltage characteristics are used as inputs to the neural network, and the initial parameters of the battery model calculated by the Nelder-Mead algorithm can be used as supervision data to calculate five initial parameters, and the SOC can be estimated with high accuracy by performing the kalman filter process.

The present invention also provides a structure of a state of charge estimating device for an electric storage device, the structure including: a measuring unit; a storage unit; the estimation unit stores data obtained by the optimization algorithm as the monitor data based on the data, and determines the initial parameter, and the calculation unit estimates the SOC by Kalman filtering processing using the initial parameter.

In the above configuration, the estimating unit includes a neural network. The neural network processing is performed using the data of the storage unit. In addition, in the above structure, the optimization algorithm uses a Nelder-Mead algorithm.

In the present specification, a neural network (also referred to as an artificial neural network) means a neural circuit network simulating living things, and has all models capable of solving problems by determining the connection strength between neurons through learning. The neural network includes an input layer, an intermediate layer (also referred to as a hidden layer), and an output layer.

In the present specification, when describing a neural network, an operation of determining the connection strength (also referred to as a weight coefficient) between neurons based on existing information is sometimes referred to as "learning".

In this specification, the work of constructing a neural network using the connection strength obtained by learning and deriving a new conclusion therefrom is sometimes referred to as "inference".

The lithium ion secondary battery is described as an example, but the present invention can be applied to other batteries (for example, all-solid-state batteries, etc.). The present invention can estimate the SOC with high accuracy by appropriately changing the battery model according to the type of the battery.

In the present specification, the state of charge (SOC) is expressed as a percentage of the sum of the remaining amount of capacity and the charged amount at the time of full charge of the secondary battery. In order to calculate the charging rate, the charging power is required to be calculated according to the pulse number, the current value and the on duty ratio of the charging current in each short time.

Effects of the invention

When the kalman filter process is performed to estimate the SOC, a neural network is used that uses data obtained by the optimization algorithm as the supervision data. By using this neural network, SOC can be estimated with high accuracy. According to one embodiment of the present invention, the SOC can be estimated with high accuracy with a relatively small calculation amount.

Drawings

Fig. 1 is an example of a block diagram showing an embodiment of the present invention.

Fig. 2 is an example of a flowchart showing an embodiment of the present invention.

Fig. 3 is an example of a flowchart showing an embodiment of the present invention.

Fig. 4 shows an example of a battery model according to an embodiment of the present invention.

Fig. 5 is a graph showing actual charge and discharge.

FIG. 6 is a graph showing the relationship between the total capacity FCC and the number of cycles.

Fig. 7 is a graph showing a relationship between the dc resistance R _s and the number of cycles.

Fig. 8 is a graph showing a relationship between the diffusion capacity C _d and the number of cycles.

Fig. 9 is a graph showing the relationship between the resistance R _d and the number of cycles generated by the diffusion process.

Fig. 10 is a graph showing the relationship between the initial SOC (0) and the number of cycles.

Fig. 11 (a) and (B) are perspective views showing an example of a secondary battery, and (C) is a schematic cross-sectional view of the secondary battery.

Fig. 12 is a diagram showing an example of a moving body.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It is noted that the present invention is not limited to the following description, and one of ordinary skill in the art can easily understand the fact that the manner and details thereof can be changed into various forms. The present invention should not be construed as being limited to the embodiments described below.

(Embodiment 1)

In this embodiment, an example in which the battery state estimating apparatus is applied to an Electric Vehicle (EV) is shown with reference to fig. 1A.

The electric vehicle is provided with a first battery 301 serving as a primary driving secondary battery and a second battery 311 supplying electric power to an inverter 312 that starts an engine 304. In the present embodiment, the state estimating unit 300 driven by the power supply of the second battery 311 uniformly monitors the plurality of secondary batteries constituting the first battery 301. The state estimation unit 300 performs state of charge estimation.

The state estimation unit 300 includes a computer including CPU (Central Processing Unit), a memory as a storage unit, and the like as main components. The CPU includes an arithmetic section that can correspond to a plurality of secondary batteries. The calculation unit determines a battery model of the secondary battery, and estimates a numerical value using a neural network. The memory stores the supervision data, and estimates the SOC from the inputted current value or voltage value.

Fig. 2 shows an example of a flowchart for estimating an SOC. The series of processes shown in fig. 2 includes the steps of: a step 1 (S1) of determining a circuit model of the secondary battery; a step 2 (S2) of storing the measured current value or the measured voltage value of the secondary battery in a storage unit; a step 3 (S3) of inputting the measured current value or the measured voltage value into a neural network using five initial parameter groups as supervision data, and a step 4 (S4) of calculating five initial parameters (FCC, R _s(R₀)、R_d、C_d, initial SOC (0)). Here, if there is an abnormal value in the initial parameters in step 4, after step 5 (S5) of changing the parameters indicating the abnormal value to values having no abnormal range (for example, the previous calculation result, etc.), step 6 of estimating the SOC by the UKF is performed (S6). For example, the abnormal value may be detected due to a change in the ambient temperature of the secondary battery, or the abnormal value may be detected based on whether or not the abnormal value is present in the initial parameter in step 4.

The monitoring data is a data group obtained by obtaining charge/discharge characteristics of several secondary batteries in actual measurement in advance and calculating five parameters using an optimization algorithm (in the present embodiment, a Nelder-Mead algorithm). When the data measured in advance is used, the learning of the state estimation unit 300 is referred to as initial learning. In addition, when the state estimation unit 300 learns data after repeating charge and discharge of the secondary battery a plurality of times and deteriorating to some extent, that is, when adding or updating the supervision data, it is also called relearning.

Fig. 3A shows an example of a learning flow. The series of processes shown in fig. 3A includes the steps of: a step 1 (S1) of determining a circuit model of the secondary battery; step 2 of measuring charge-discharge characteristics of the secondary battery (S2); step 3 (S3) of calculating five initial parameters by optimizing the charge-discharge characteristics of the secondary battery by a Nelder-Mead algorithm; and a step 4 of constructing a neural network for learning the five initial parameter groups as the supervision data (S4).

All or part of the above-described respective processes may be set as an automatic operation step. Further, a part of the steps may be performed manually at the timing of the user, or may be performed periodically. The order of steps shown in the figures may be arbitrarily changed, as well as information comprising various data or parameters. That is, the information shown in the drawings is of course not limited to the order shown.

The monitoring data may be charge/discharge characteristics of the nth cycle (n is an integer of 2 or more) obtained in advance.

Fig. 3B shows an example of a learning flow. The series of processes shown in fig. 3B includes: a step 1 (S1) of determining a circuit model of the secondary battery; step 2 of measuring charge-discharge characteristics of the secondary battery (S2); and step 3 of calculating five initial parameters by optimizing the charge and discharge characteristics of the secondary battery by a Nelder-Mead algorithm (S3). If the actual measured cycle data can be obtained in advance, the method further comprises a step 15 of calculating five initial parameter groups for each n cycles by optimizing the actual measured data of the secondary battery for each n cycles (S15) and a step 17 of constructing a neural network for learning the five initial parameter groups as supervision data (S17).

Further, a part of the supervision data may not be actually measured, but may use charge and discharge characteristics assumed by the practitioner. In this case, if the actually measured cycle data cannot be acquired in advance after step 3, step 16 of setting a virtual initial parameter group for each condition of the five initial parameters may be performed before step 17 (S16). This step 16 may be referred to as an additional learning step. The virtual initial parameter group may also be referred to as virtual supervision data. Further, since the supervision data is added or updated when the initial parameter group is changed, the method may be called relearning.

The first battery 301 mainly supplies electric power to the 42V series (high voltage series) vehicle-mounted device, and the second battery 311 supplies electric power to the 14V series (low voltage series) vehicle-mounted device. As the second battery 311, a lead storage battery is used in many cases because of cost advantages. Lead-acid batteries have the disadvantage of being large in self-discharge compared to lithium-ion secondary batteries and being susceptible to degradation by a phenomenon known as sulfation. Although there is an advantage in that maintenance is not required when the lithium ion secondary battery is used as the second battery 311, an abnormality which cannot be distinguished at the time of manufacture may occur during a long period of use, for example, three years or more. Particularly if the second battery 311, which starts the inverter, is not operated, the generator cannot be started even if the first battery 301 has a remaining capacity. In order to prevent this, in the case where the second battery 311 is a lead storage battery, electric power is supplied from the first battery to the second battery and charging is performed so as to maintain a fully charged state at all times.

The present embodiment shows an example in which both the first battery 301 and the second battery 311 use lithium ion secondary batteries. The second battery 311 may also use a lead storage battery or an all-solid-state battery.

The regenerative energy generated by the rotation of the tire 316 is transmitted to the engine 304 through the transmission 305, and is supplied to the second battery 311 or the first battery 301 through the engine controller 303 and the battery controller 302.

The first battery 301 is mainly used to rotate the engine 304, and also supplies electric power to 42V-series vehicle-mounted components (an electric power steering system 307, a heater 308, a defogger 309, and the like) through a DCDC circuit 306. The first battery 301 is used to rotate the rear engine in the case where the rear wheel includes the rear engine.

The second battery 311 supplies electric power to 14V series vehicle-mounted components (acoustic apparatus 313, power window 314, lamps 315, etc.) through the DCDC circuit 310.

The first battery 301 is constituted by a plurality of secondary batteries. For example, a cylindrical secondary battery 600 is used. As shown in fig. 1B, a module 615 may be formed by sandwiching a cylindrical secondary battery 600 between a conductive plate 613 and a conductive plate 614. The switch between the secondary batteries is not illustrated in fig. 1B. The plurality of secondary batteries 600 may be connected in parallel, may be connected in series, or may be connected in series after being connected in parallel. By constructing the module 615 to include a plurality of secondary batteries 600, greater power may be extracted.

In fig. 1A, the battery controller 302 and the state estimation unit 300 are shown as separate structures, but there is no particular limitation, and they may be formed of one IC chip on the same substrate, or may be combined into one unit. The state estimation unit 300 may be constituted by an LSI (LARGE SCALE Integration) manufactured integrally on one chip. The method of integrating the circuit is not limited to LSI, and may be realized by a dedicated circuit or a general-purpose processor. Further, an FPGA (Field Programmable GATE ARRAY) programmable after LSI manufacturing or a reconfigurable processor capable of reconfiguring connection or setting of circuit cells inside the LSI may be used. In addition, an IC (also referred to as an inference chip) in which an AI system is assembled may also be used. The IC in which the AI system is assembled is sometimes called a circuit (microcomputer) that performs a neural network operation. In addition, the battery controller 302 is sometimes referred to as BMU (Buttery Management Unit). The five initial parameters are stored in, for example, a memory of the state estimation unit 300 of the secondary battery, specifically ROM (Read Only Memory) or RAM (Random Access Memory). The state estimating unit 300 can calculate the SOC of the secondary battery more accurately.

A power storage control device or a management device including the state estimation unit 300 of the secondary battery may be realized. Further, a power storage control method may be realized in which a plurality of processes including a neural network process are sequentially performed, constituting a plurality of steps. The steps included in the power storage control method may be implemented as a computer program executed by a computer. Such a computer program may be stored and executed on a recording medium or a cloud of a communication network using the internet.

Programs of software for executing the computer program may be written in various programming languages such as Python, go, perl, ruby, prolog, visual Basic, C, C ++, swift, java (registered trademark), NET, and the like. The application may be written using frameworks such as Chainer (available in Python), caffe (available in Python and c++), tensorFlow (available in C, C ++ and Python).

(Embodiment 2)

Fig. 4B shows an example of a battery model.

The battery model shown in fig. 4B in the present embodiment is a model in which the model shown in fig. 4A is simplified. The weber impedance portion is infinite, but fifty stages are shown simplified in fig. 4A. In the weber impedance portion of fifty stages shown in fig. 4A, the fourth stage to the fifty stage are set as resistors, and the cells having small time constants are summarized as the direct current resistor Rs in fig. 4B. Fig. 4B shows a series connection of a dc resistance model and a diffusion resistance model.

The resistance R _d generated by the diffusion process represents the resistive component, and the diffusion capacity C _d represents the capacity component term. The diffusion resistor is a parallel connection body of a resistance component and a capacity component, and is configured by connecting a plurality of parallel connection bodies in series (three stages in the drawing). The equivalent circuit formed by the parallel connection of the resistive component and the capacity is called a foster circuit model. The foster type circuit model is preferable because it is less computationally intensive than the cole type circuit model.

In the simplified model of fig. 4, three parameters (R _s、R_d、C_d) can be used.

Further, the OCV may be expressed as follows.

[ Formula 1]

OCV＝f(SOC(t))

Further, the SOC (t) may be expressed as follows.

[ Formula 2]

Further, the state variable x (t) may be expressed as follows.

[ Arithmetic 3]

x(t)＝[SOC(t) v₁(t) v₂(t) v₃(t)]

Further, the output equation may be expressed as follows.

[ Calculation formula 4]

V(t)＝f(SOC(t))+v₁(t)+v₂(t)+v₃(t)+R_sI(t)

Therefore, if these expressions can be used to find five initial parameters, the calculation can be performed from the state space.

[ Calculation formula 5]

[FCC Rs Rd Cd SOC(0)]

In the present embodiment, these five initial parameters are calculated from the actual measurement data of the voltage and the current by optimization. Although there is a least square method as an optimization algorithm, since the secondary battery has a nonlinear characteristic, a Nelder-Mead algorithm is used as the optimization algorithm. Here, as an example, five initial parameters are calculated by optimizing using the data of the current shown in fig. 5A and the data of the voltage shown in fig. 5B.

A neural network is constructed that learns one of the supervised data with five initial parameter values.

To verify the use of cyclic test data. The conditions for the cyclic test data used were: the ambient temperature was 45 ℃, the charge cut-off voltage was 4.2V, the discharge cut-off current was 2.5V, the charging regime was CC-CV, the charging rate in CC was 0.5C (1.625A), and the discharging rate was 1C (3.25V).

Five initial parameters obtained by optimizing the data for each cycle are shown as black lines for comparison. Fig. 6, 7, 8, 9, and 10 show the calculated values, respectively.

The input layers are 700, the first hidden layers are 500, the second hidden layers are 500, and the output layers are five (FCC, R _s、R_d、C_d and initial SOC (0)) by adopting the fully-connected neural network. When the hidden layers of the neural network overlap, this is also called deep learning. In the present embodiment, although an example in which a fully-connected neural network is used is shown, the structure of the neural network or the learning method is not particularly limited.

The neural network processing was also performed in the fifty th cycle, 150 th cycle, 250 th cycle, 350 th cycle, 450 th cycle, 550 th cycle, and 650 th cycle to calculate five parameters, and each of the data is shown by circles in fig. 6, 7, 8, 9, and 10.

Using the initial parameters obtained by learning the neural network of the supervised data obtained by performing the optimization algorithm, it is possible to obtain substantially the same values as the data actually subjected to the optimization. When the optimization algorithm is used, there is a possibility that unnecessary iterative processing, convergence to a physically meaningless value, or divergence problems occur, and it is difficult to use only the optimization algorithm for the estimation, but the neural network processing using the monitor data that has been optimized can calculate the initial parameters.

Accordingly, it can be said that the initial parameters calculated by the neural network processing using the optimized monitor data are appropriate values as initial parameters for the kalman filter, and therefore the accuracy of the SOC obtained by the kalman filter processing is improved.

Embodiment 3

An example of a cylindrical secondary battery is described with reference to fig. 11A and 11B. As shown in fig. 11A, the top surface of the cylindrical secondary battery 600 includes a positive electrode cap (battery cap) 601, and the side and bottom surfaces thereof include a battery can (outer can) 602. These positive electrode cover 601 and battery can (outer can) 602 are insulated by a gasket (insulating gasket) 610.

Fig. 11B is a view schematically showing a cross section of a cylindrical secondary battery. A battery element in which a band-shaped positive electrode 604 and a band-shaped negative electrode 606 are wound with a separator 605 interposed therebetween is provided inside a hollow cylindrical battery can 602. Although not shown, the battery element is wound around the center pin. One end of the battery can 602 is closed and the other end is open. As the battery can 602, a metal having corrosion resistance to an electrolyte, such as nickel, aluminum, titanium, or the like, an alloy thereof, or an alloy thereof with other metals (e.g., stainless steel, or the like) may be used. In addition, in order to prevent corrosion by the electrolyte, the battery can 602 is preferably covered with nickel, aluminum, or the like. Inside the battery can 602, a battery element in which the positive electrode, the negative electrode, and the separator are wound is sandwiched between a pair of insulating plates 608 and 609 that face each other. A nonaqueous electrolyte (not shown) is injected into the battery can 602 in which the battery element is provided. The secondary battery is composed of a positive electrode including an active material such as lithium cobalt oxide (LiCoO ₂) or lithium iron phosphate (LiFePO ₄), a negative electrode made of a carbon material such as graphite capable of occluding and releasing lithium ions, and a nonaqueous electrolyte solution in which an electrolyte composed of a lithium salt such as LiBF ₄、LiPF₆ is dissolved in an organic solvent such as ethylene carbonate or diethyl carbonate.

Since the positive electrode and the negative electrode for the cylindrical secondary battery are wound, the active material is preferably formed on both surfaces of the current collector. Positive electrode 604 is connected to positive electrode terminal (positive electrode collector lead) 603, and negative electrode 606 is connected to negative electrode terminal (negative electrode collector lead) 607. As the positive electrode terminal 603 and the negative electrode terminal 607, a metal material such as aluminum can be used. The positive terminal 603 is resistance welded to the safety valve mechanism 612 and the negative terminal 607 is resistance welded to the bottom of the battery can 602. The safety valve mechanism 612 is electrically connected to the positive electrode cover 601 via a PTC (Positive Temperature Coefficient: positive temperature coefficient) element 611. When the internal pressure of the battery rises above a predetermined threshold value, the safety valve mechanism 612 cuts off the electrical connection between the positive electrode cover 601 and the positive electrode 604. In addition, the PTC element 611 is a thermosensitive resistor whose resistance increases when the temperature rises, and limits the amount of current by the increase in resistance to prevent abnormal heat generation. As the PTC element, barium titanate (BaTiO ₃) semiconductor ceramics or the like can be used.

A lithium ion secondary battery using an electrolyte solution includes a positive electrode, a negative electrode, a separator, an electrolyte solution, and an exterior body. Note that in a lithium ion secondary battery, since the anode and the cathode are changed over by charge or discharge, the oxidation reaction and the reduction reaction are changed over, and therefore, an electrode having a high reaction potential is referred to as a positive electrode, and an electrode having a low reaction potential is referred to as a negative electrode. In this way, in the present specification, even when charge, discharge, reverse pulse current flow, and charge current flow, the positive electrode is referred to as "positive electrode" or "+electrode", and the negative electrode is referred to as "negative electrode" or "+electrode". If the terms of anode and cathode are used in connection with oxidation and reduction reactions, the anode and cathode are reversed when charged and discharged, which may cause confusion. Therefore, in the present specification, the terms anode and cathode are not used. When the terms anode and cathode are used, it clearly indicates whether it is charged or discharged, and shows whether it corresponds to a positive electrode (+electrode) or a negative electrode (-electrode).

Two terminals shown in fig. 11C are connected to a charger to charge the battery 1400. As the charge of the battery 1400 progresses, the potential difference between the electrodes increases. The positive direction in fig. 11C is: from the terminal outside the battery 1400, to the positive electrode 1402, from the positive electrode 1402 to the negative electrode 1404 in the battery 1400, and from the negative electrode to the terminal outside the battery 1400. That is, the direction in which the charging current flows is the direction of the current.

In the present embodiment, an example of a lithium ion secondary battery is shown, but is not limited to a lithium ion secondary battery. As the positive electrode material of the secondary battery, for example, a material containing element a, element X, and oxygen can be used. The element a is preferably one or more elements selected from the group consisting of a first group element and a second group element. As the first group element, for example, alkali metals such as lithium, sodium, and potassium can be used. As the second group element, for example, calcium, beryllium, magnesium, or the like can be used. As the element X, for example, one or more elements selected from the group consisting of metal elements, silicon, and phosphorus can be used. The element X is preferably one or more elements selected from cobalt, nickel, manganese, iron, and vanadium. Typically, lithium cobalt composite oxide (LiCoO ₂) and lithium iron phosphate (LiFePO ₄) are cited.

The anode includes an anode active material layer and an anode current collector. The negative electrode active material layer may contain a conductive auxiliary agent and a binder.

As the negative electrode active material, an element that can undergo a charge-discharge reaction by an alloying/dealloying reaction with lithium can be used. For example, a material containing at least one of silicon, tin, gallium, aluminum, germanium, lead, antimony, bismuth, silver, zinc, cadmium, indium, and the like can be used. The capacity of this element is greater than that of carbon, especially silicon, by 4200mAh/g.

In addition, the secondary battery preferably includes a separator. As the separator, for example, a separator formed of a fiber having cellulose such as paper, a nonwoven fabric, a glass fiber, a ceramic, a synthetic fiber including nylon (polyamide), vinylon (polyvinyl alcohol fiber), polyester, acrylic resin, polyolefin, polyurethane, or the like can be used.

Fig. 12 illustrates a vehicle using a state of charge estimating device for a secondary battery according to an embodiment of the present invention. The secondary battery 8024 of the automobile 8400 shown in fig. 12A may supply electric power to a light emitting device such as a headlight 8401 or an indoor lamp (not shown) in addition to the motor 8406. As the secondary battery 8024 of the automobile 8400, a module 615 in which the cylindrical secondary battery 600 shown in fig. 11B is sandwiched between the conductive plate 613 and the conductive plate 614 may be used.

In the automobile 8500 shown in fig. 12B, the secondary battery of the automobile 8500 can be charged by receiving electric power from an external charging device by a plug-in system, a contactless power supply system, or the like. Fig. 12B shows a case where a secondary battery 8024 mounted in an automobile 8500 is charged from a charging device 8021 provided on the ground via a cable 8022. In the case of charging, the charging method, the specification of the connector, and the like may be appropriately performed according to a predetermined system such as CHAdeMO (registered trademark) and a combined charging system. As the charging device 8021, a charging station provided in a commercial facility or a power supply in a home may be used. For example, by supplying electric power from the outside using the plug-in technology, the secondary battery 8024 mounted in the automobile 8500 can be charged. The charging may be performed by converting AC power into DC power by a conversion device such as an AC/DC converter.

Although not shown, the power receiving device may be mounted in a vehicle and may be charged by supplying electric power from a power transmitting device on the ground in a noncontact manner. When the noncontact power feeding method is used, the power transmission device is assembled to the road or the outer wall, so that charging can be performed not only during the stop but also during the traveling. Further, the noncontact power feeding method may be used to transmit and receive electric power between vehicles. Further, a solar cell may be provided outside the vehicle, and the secondary battery may be charged during parking or traveling. Such non-contact power supply can be realized by electromagnetic induction or magnetic field resonance.

Fig. 12C is an example of a two-wheeled vehicle using a secondary battery according to an embodiment of the present invention. The scooter 8600 shown in fig. 12C includes a secondary battery 8602, a rear view mirror 8601, and a turn signal 8603. The secondary battery 8602 may supply power to the directional lamp 8603.

In the scooter type motorcycle 8600 shown in fig. 12C, the secondary battery 8602 may be stored in an under-seat storage box 8604. Even if the under-seat storage box 8604 is small, the secondary battery 8602 can be stored in the under-seat storage box 8604.

This embodiment mode can be appropriately combined with the description of other embodiment modes.

[ Description of the symbols ]

300: State estimation section 301: battery, 302: battery controller, 303: engine controller, 304: engine, 305: transmission, 306: DCDC circuit, 307: electric power steering system, 308: heater, 309: demister, 310: DCDC circuit, 311: battery, 312: inverter, 313: acoustic equipment, 314: power window, 315: lamps, 316: tire, 600: secondary battery, 601: positive electrode cap, 602: battery can, 603: positive electrode terminal, 604: positive electrode, 605: separator, 606: negative electrode, 607: negative electrode terminal, 608: insulation board, 609: insulation board, 611: PTC element, 612: safety valve mechanism, 613: conductive plate, 614: conductive plate, 615: module, 1400: storage battery, 1402: positive electrode, 1404: negative electrode, 8021: charging device, 8022: cable, 8024: secondary battery, 8400: automobile, 8401: headlight, 8406: electric motor, 8500: automobile, 8600: scooter, 8601: rearview mirror, 8602: secondary battery, 8603: direction light, 8604: and a storage box under the seat.

Claims

1. A charge state estimation method of an electric storage device, comprising the steps of:

determining a circuit model of the power storage device;

In a circuit model of the power storage device, a current is taken as an input and a voltage is taken as an output;

Optimizing the output error of the voltage of the power storage device to be small, and calculating a first initial parameter value of a circuit model of the power storage device;

storing initial parameter groups corresponding to input values of different currents; and

And determining a second initial parameter value by neural network processing using the initial parameter group as the supervision data, and estimating a charging rate by kalman filter processing using the second initial parameter value for the current value or the voltage value of the power storage device.

2. A charge state estimation method of an electric storage device, comprising the steps of:

determining a circuit model of the power storage device;

Generating a virtual initial parameter group calculated by using the assumed charge-discharge characteristics; and

3. The charge state estimation method of an electrical storage device according to claim 1 or 2,

Wherein the first initial parameter value is at least one of FCC, R _s、R_d、C_d, and initial SOC (0).

4. A state of charge estimating device of an electric storage device, comprising:

a measuring unit;

A storage unit;

A presumption part; and

An operation unit for performing an operation on the data,

Wherein the measuring unit measures the current or voltage of the power storage device,

The storage unit stores the data measured by the measuring unit,

The presumption part includes a neural network using a data group obtained by an optimization algorithm from the data as supervisory data,

The estimation unit determines an initial parameter value using the supervision data,

The calculation unit estimates the charging rate by performing a kalman filter process on the current value or the voltage value of the power storage device using the initial parameter value.

5. The charge state estimation device of an electrical storage device according to claim 4,

Wherein the optimization algorithm uses a Nelder-Mead algorithm.